Greysight

An open source Snowflake cost optimization and observability dashboard.

We're team Greybeam, we work with many Snowflake users who often have very narrow visibility into their Snowflake costs. Many solely rely on the Cost Management tab provided by Snowflake and that's obviously never enough.

Snowflake tells you what you spent. Finding out where it went usually means writing your own queries against ACCOUNT_USAGE and ORGANIZATION_USAGE and rebuilding the same charts every team builds. Greysight does that part for you: it reads approved, read-only metadata queries, turns them into a cost dashboard, and renders it in a Next.js app. Greysight also includes Auto Savings, a feature that suspends warehouses when Snowflake metadata proves they are idle.

We offer a free hosted version at costs.greybeam.ai. In case you want more info before signing up you can find that here. Or run it yourself: using the quick start below.

Features

• Cost observability: Attribute Snowflake spend, inspect warehouse and query costs, and review usage trends from one dashboard.
• Auto Savings: Reduce idle compute by suspending active warehouses that have no running queries.

Quick start

You need Node.js 20+, npm, and uv with Python 3.12.

Run Greysight locally with a Snowflake account

Copy .env.example to .env, set DATA_SOURCE=snowflake, keep AUTH_REQUIRED=false, and fill in your connection:

DATA_SOURCE=snowflake
AUTH_REQUIRED=false
NEXT_PUBLIC_API_BASE_URL=http://localhost:8000
SNOWFLAKE_ACCOUNT=
SNOWFLAKE_USER=
SNOWFLAKE_ROLE=
SNOWFLAKE_WAREHOUSE=
SNOWFLAKE_DATABASE=SNOWFLAKE
SNOWFLAKE_SCHEMA=ACCOUNT_USAGE
SNOWFLAKE_PRIVATE_KEY_PATH=/absolute/path/to/key.p8
SNOWFLAKE_PRIVATE_KEY_PASSPHRASE=

The role you authenticate with needs read access to the ACCOUNT_USAGE and ORGANIZATION_USAGE views. Then restart npm run dev. Full walkthrough: docs/snowflake-setup.md.

Greysight only uses the read-only SQL at sql/snowflake and approved in sql/dashboard_sources.yml for dashboard data. This path does not write to your account, and Snowflake credentials are not returned to the browser. The separately enabled Auto Savings worker is the only exception: it can issue ALTER WAREHOUSE … SUSPEND for warehouses that you explicitly enroll.

How Greysight works

The data path is intentionally boring:

1. Read-only SQL lives in sql/snowflake.
2. Each dataset is registered in sql/dashboard_sources.yml.
3. FastAPI runs the approved sources (or loads demo data).
4. The backend computes the metrics and builds a prepared dashboard view.
5. The Next.js app fetches, validates, caches, and renders that view.

The frontend never invents its own analytics. If a chart needs a new derived number, it gets added to the backend view contract first, so the dashboard and the data always agree.

Auto Savings

Auto Savings runs as a separate worker. An organization must agree to use the feature, enable its organization-wide switch, and enroll and enable each warehouse before the worker can act on it.

For each organization, the worker:

1. Reads the current warehouse state with SHOW WAREHOUSES.
2. Checks each enabled enrollment against that snapshot.
3. Confirms the warehouse is a running STANDARD warehouse with no running or queued statements, no quiescing compute, AUTO_RESUME enabled, and at least 60 seconds of uptime.
4. Rechecks the current switches, warehouse identity, and enrollment version.
5. Issues ALTER WAREHOUSE … SUSPEND.

The full eligibility and fail-closed behavior is described in How Auto Savings works.

Try Auto Savings locally

Auto Savings currently requires a Supabase backend. A fully local SQLite or DuckDB backend is planned for a future release.

1. Follow the Supabase setup, apply the migrations in supabase/migrations, and copy .env.example to .env.

2. Configure the root .env for authenticated Snowflake mode. SUPABASE_URL must be set explicitly because the worker does not read the browser-facing URL as a fallback.

   DATA_SOURCE=snowflake
   AUTH_REQUIRED=true
   NEXT_PUBLIC_AUTH_REQUIRED=true
   NEXT_PUBLIC_API_BASE_URL=http://localhost:8000
   NEXT_PUBLIC_SUPABASE_URL=
   NEXT_PUBLIC_SUPABASE_ANON_KEY=
   SUPABASE_URL=
   SUPABASE_SERVICE_ROLE_KEY=

3. Run npm run dev, sign in, and connect the organization to Snowflake using the in-app connection flow.

4. Give the configured Snowflake role permission to suspend the test warehouse. The Autod Savings operations guide describes the supported account-level and per-warehouse grants.

5. Open Auto Savings in Greysight, agree to the feature, enroll the test warehouse, and enable both the organization and warehouse switches.

6. Start the worker in a second terminal:

   cd apps/auto-savings
   uv run dev.py

7. Resume the test warehouse and leave it idle. After the warehouse has been running for at least 60 seconds, the worker can suspend it on an eligible polling cycle. Worker decisions and command outcomes are written as structured logs.

Project layout

apps/web/                Next.js dashboard, auth/org UI, browser API clients, tests
apps/api/                FastAPI backend, Snowflake access, metrics, route tests
apps/auto-savings/       Opt-in warehouse suspension worker and tests
shared/connect/          Shared Snowflake and Supabase connection package
sql/snowflake/           Approved read-only Snowflake SQL assets
sql/dashboard_sources.yml  Dataset registry for dashboard sources
supabase/migrations/     Schema, RLS, org membership, credential storage
docs/                    Setup, Snowflake, security, and deployment notes

Development

From the repository root:

npm run dev          # web :3000 + API :8000
npm run test         # Vitest + pytest
npm run lint         # ESLint + ruff
npm run typecheck    # TypeScript

The test suites are hermetic, they never call Snowflake or Supabase, and demo mode stays fully local, so you can work on the dashboard without any external credentials.

Docs

• Local development
• Snowflake setup
• How Auto Savings works
• Auto Savings operations guide
• Security model
• Deployment
• Dependency compatibility

Contributing

Greysight is young and 0.0.1 is intentionally narrow — it's a cost dashboard, not a recommendations engine (yet). Small, focused changes are easiest to review.

A typical dashboard change touches the SQL source or backend metric, the prepared view contract, the frontend view, the demo data, and tests. Keep analytics in the backend and presentation in the frontend, keep secrets out of the browser, and let Supabase own auth while the backend owns authorization and Snowflake access.

License

Apache-2.0.